We do not predict markets.
We study them.
Every forecast on this site is a frozen experiment. Every result is measured against reality. Every error is published below, permanently. The objective is not to be right. The objective is to become incrementally less wrong, every day.
- A trading bot
- A signal service
- Financial advice
- A prediction engine
- A scientific research platform
- An evidence engine
- A probability laboratory
- A public, auditable record
Calibration, measured. Not claimed.
When we state 80% confidence, reality should agree close to 80% of the time. Higher is not better. Closer to target is better. The distance is the score.
Actual is the pooled share of all resolved experiments whose closing price landed inside its frozen 80 percent interval. Score % is the Winkler interval score divided by the interval midpoint, so every asset is on one scale: for context, a perfectly calibrated 80 percent interval on these assets currently spans roughly 1 to 2 percent of price. * Fewer than 30 resolved experiments: results preliminary.
Null honesty measures whether the AI correctly concludes nothing notable happened on quiet windows instead of inventing a story. An AI that confabulates on noise confabulates everywhere.
| Asset | Resolved | Actual | Target | Score % | Trend |
|---|
Hit rate is judged by distance from the stated target, not by height. Green means within 10 points of target; amber means the method is miscalibrated for that asset and we say so.
Experiment Log
Hypotheses and forecast intervals are frozen and hashed before each observation window opens. What you see is what was committed, including every miss and every void.
Every article traces to real experiments.
One entry per day: yesterday's forecast versus reality, the largest surprise, what the evidence changed, and tomorrow's research question. Written over a frozen manifest, never from imagination.
One documented rule, watching in real time.
Every 15 minutes this rule checks the live frozen RENDER and TAO forecasts against the current price and decides: hold, or place a small limit order. The rule is fixed and fully stated below; nothing here is discretionary, and nothing here is a recommendation to you.
A transparent research and development validation running against a small, fixed R&D budget on a real exchange account. Two separate signals, each doing only its own job: a longer-horizon forecast first answers whether an asset has real predicted upside left at all (a strategic filter, currently requiring price to sit 4%+ below the predicted 24-hour high before a buy is even considered); only then does the shorter-horizon frozen band set the actual entry price, sized $5–$15 by how far price has breached the band's edge. Every position is tracked with a real cost basis from the moment it opens. Exits are gated the same way: a take-profit only fires once real profit clears fees and a minimum edge, scaling the exit size with how much profit has run; a full close only fires on a genuine 5%+ move against the tracked entry, not a routine band touch. One position per asset at a time. This is not financial advice, not a signal service, and not a recommendation to trade. It exists to answer one engineering question: does CryptoIQ's own calibrated forecast produce anything usable once real fees, spreads, and execution are accounted for.
"Managed crypto value" covers only actively-traded symbols. The Kraken comparison includes every asset on the account, including small legacy/dust holdings outside the managed symbols -- a nonzero gap here is expected; what matters is whether it drifts unexpectedly over time, not whether it's exactly zero.
Forecast improves by replaying history it never saw.
Each run picks a random historical window, then steps forward one candle at a time using only data available at that instant, no lookahead. It runs the exact same forecasting method production uses. When the run ends, an AI critiques the results and proposes improvements, ranked and reasoned. It cannot change anything itself; every proposal sits in a queue for a human to accept or reject. Nothing here executes trades or touches the live account.
A separate, newer kind of test: instead of measuring forecast calibration, this replays the ACTUAL live trading rule against real historical hourly bands, real 24-hour forecasts (whichever was genuinely live at each hour, no lookahead), and real candle prices, including real Kraken fees. It answers a different question than the replay above: not "does the forecast calibrate," but "would this specific trading rule have made or lost real money." Read-only measurement; nothing here executes trades.
Individual simulated trades
Expanding the universe, one verified stage at a time.
Scout never trades, never forecasts, never skips a stage, and never tells you what to buy. Every candidate moves universe → candidate → forecasting → research → strategy one step at a time, scored on tradability, data sufficiency, volatility, diversification against the active portfolio, and forecastability, whether the current method actually calibrates on this asset's real history. Tap any candidate for everything we've measured on it.
Confirmed picks, validated by one real week of simulated trading
Composite score alone isn't enough. Each candidate below was actually walked forward through the real Strategy Lab rule against its most recent real week, zero lookahead, real Kraken fees deducted. Only candidates with at least one completed trade and a net positive result count as confirmed; the ranked list often has to go deep before finding five.
Full ranked list, pre-validation composite score
Every candidate is rescored from scratch nightly against the entire tradable universe. This ranking alone does NOT mean a candidate would trade profitably, see Confirmed Picks above for the ones actually tested.
Surprises
Every measurement Scout Lab has recorded for this candidate, plainly stated. This is evidence, not a recommendation. Nothing here is investment advice, and nothing here tells you what to do with it.
Dimension scores
Raw metrics
Proposal record
Surprises involving this candidate
Lookback search history
One-week trading simulation
The CryptoIQ Constitution
- Article I. Facts are immutable.
- Article II. All measurements are reproducible.
- Article III. Forecasts are frozen before observation begins.
- Article IV. Evaluations are computed, never generated.
- Article V. AI may explain, but never invent, evidence.
- Article VI. Confidence is earned statistically.
- Article VII. Every conclusion must be auditable.
- Article VIII. The objective is reducing uncertainty, not maximizing profit.
- Article IX. Absence of evidence is evidence of uncertainty, not certainty of absence.
These articles are not policy. They are compiled into database permissions: fact tables accept inserts and refuse all edits, forecasts freeze atomically with their hypotheses, and confidence values can only be written by statistical computation. The philosophy is enforced, not promised.
Powered by the X-Vault Framework
X-Vault has always done one thing: transform massive, complex information into evidence-based understanding. Government archives at DeClassX.com. Ancient manuscripts at SmootCodex.org. Now, financial markets.
CryptoIQ.trade applies the scientific method to markets. It combines market data, technical measurements, macroeconomic events, and AI-assisted research to help investors understand why markets move, not just that they moved. Every AI-generated sentence traces to the exact facts it was shown. Every number is computed, never imagined.
CryptoIQ.trade is a research and educational platform. Nothing on this site is financial advice, an offer, or a recommendation to buy or sell any asset. Forecasts are statistical experiments about our own calibration, not guidance. Markets involve substantial risk. All investment decisions are yours alone.